Software Issues 6: Application Architectures: Past, Present, and Future
Back in the "olden days," when relational databases first started gaining widespread use, we wrote standalone front-end applications. Although we didn't know it at the time, we were writing single-tier applications. What this term means is that all the code is in a single application, including the user interface logic, the business rules, and the database management.
As these applications gained in popularity, it became evident that they were too difficult to maintain. Large companies might have dozens or even hundreds of users running the applications. Every time a business rule changed, the program had to be modified, recompiled, and redistributed to all those people. At that time, that meant some pitiful slob walking around the building running installs, reconfiguring, and troubleshooting, while all the time the users' work was interrupted. It got to be intolerable.
To mitigate this problem, efforts shifted to two-tier applications. By expanding the capabilities of stored procedures in the database, business logic could be moved there. That meant that stored procedures could be stored and maintained in one place. If they needed to be changed, they could be modified without having to recompile and re-deploy new program versions.
This architecture helped, but soon proved too limited:
As systems started to support larger numbers of users and the complexity of stored procedures grew, the load on the database became a limiting factor.
SQL is not a programming language. As stored procedures were pushed to do procedural things normally handled by powerful programming languages, the stored procedures became difficult to write, difficult to debug, and difficult to maintain.
Portability became an issue. To move to a new database required that all the stored SQL code—perhaps the majority of the complex code in the application—had to be rewritten.
To deal with these issues, three-tier applications became the accepted standard. In a three-tier application, you basically backed away from putting all the business logic in stored procedures. Instead, you moved business logic in a programming environment such as Visual Basic, and deployed that code as middle-tier components, DLLs. These business objects could reside on the database server or on entirely separate servers. This architecture allowed the developers to distribute services and load on three separate machines. It provided the benefit of a thin front-end program, a dedicated database, and business code, written in a powerful language and maintainable in one place on the server side.
It would seem as though this strategy should have met our all our needs. However, as Internet front-end programs became more common, the number of potential users jumped once more. When that happened, even these three-tier architectures were not enough. The middle tier couldn't handle the load.
The solution was to move to n-Tier architectures. An n-Tier architecture is basically a three-tier architecture with n number of middle-tier servers. Those servers can reside at different locations. Now the workload can be distributed in various ways across those servers; for example, by business function. Also, fault tolerance can be increased by having redundant servers in the middle tier.
Of course, there is a lot more to the story, but that's the bare bones. Now we've extended it further by not only scaling out our middle tier, but also our database tier, as well as providing intelligent load balancing and other services.
So far, we've ignored the Internet, and the Internet cannot be ignored. It has had and continues to have profound impact on application architectures.
So let's look back to the influence of the Internet. One of the many reasons the Internet took off so rapidly was that browser-based applications further reduced the maintenance requirements of front-end programs, and also solved connectivity issues through the World Wide Web. By creating a front end that was browser-based rather than VB-based, there was no need to do any installation or configuration. Installation and configuration has always been a major headache, even for a thin VB client. Therefore, when you have potentially tens of thousands of users, the browser was the only solution.
For a while, it looked like VB applications would soon be extinct and every application would be browser-based in a very short time. Although browsers didn't offer the kind of power and flexibility that VB applications provide, the benefits far outweighed the limitations for most customers. Browser developers scrambled to add more capability, and the entire world basically accepted the limitations of browser-based applications as a temporary condition.
Some developers and some marketers still think that all applications today need to be browser-based. However, the technical landscape has continued to shift. One of the main motivations for browser-based applications was ease of deployment. However, in the last few years Windows has gotten better about installation and configuration. It's going to get better yet. Terminal Services and Citrix offer a very efficient alternative for some applications. Finally, Active Update programs have dramatically lowered the maintenance barriers. Communication across the Web is now becoming transparently simple for standalone applications.
One great example of using these alternate approaches to architect sophisticated Internet-enabled applications is online gaming. The game guys are always way out at the leading edge of the wave. Everquest is one example of such a game. It's a heavy-duty client/server application running across the Internet. In Everquest's 3D world, thousands of players interact in real time. This application has an extremely heavy front-end application, consisting of over half a gigabyte of data and code on the local machine alone. And all of the critical information is stored and processed on the server.
This game simply could not have been developed within a browser. There are more than 300,000 Everquest subscribers; every time they run the program, it's updated automatically over the Web. Modifications are made almost daily, and all 300,000 people routinely get updated. If the designers of Everquest can accomplish this level of maintainability and performance, virtually any intensive application can follow this same model.
This illustrates the reality that you needn't think only of browser-based applications when you think of the Internet. There are other viable options, including Terminal Services–based applications, running inside or outside of a browser; installed applications running across the Web; even Internet-installed and Internet-updated applications.
The reasons that got us thinking only about browser-based solutions have already gone away. In the near future, the Microsoft.Net platform will make it even easier to copy heavy applications to client machines over the Web—without having to perform installations—as well as facilitate Web connectivity.
Evaluate your application needs prior to deciding on an architecture. The first question to ask is whether the application needs to be Internet-enabled. If the answer is yes, ask whether it's a casual or intensive application. If it's casual, think about a browser-based solution. If it's intensive, consider a shipped or Web-deployed Internet-aware application or a Terminal Services solution.
The distinction between casual and intensive is important. It's not meant to suggest a value judgment, but rather usage patterns and expectations. Casual applications are used only occasionally; for example, Web commerce applications such as online booksellers or trading sites. Users don't use these applications intensively or as part of the job. They don't want these sites to download and install custom components on their systems, and likewise don't expect functionality beyond what's offered by the browser.
An intensive application is a business application with heavy requirements that is used by employees or clients regularly to perform more intensive operations. These may include data entry, analysis, and reporting. Clients expect sophisticated custom interfaces to help them perform tasks efficiently. Managers expect to provide tools that optimize the productivity of their workers, not necessarily to make the task feasible in a browser.
While Internet Explorer is powerful, for example, it's unlikely ever to offer the power and flexibility of a standalone application. It's tremendous for casual applications, but often requires developers and users to make significant compromises in the design of more intensive business applications. Therefore, why spend a lot of time and effort and realize marginal results trying to force intensive functionality into a browser? There are alternatives.
On the other hand, why incur the overhead of supporting installed applications when the user would rather not leave the convenience of the browser for casual operations? The browser is clearly the best choice for casual applications. Other techniques may be better for more intensive applications. Of course, there is a wide gray area within which the choice is not clear. Time to market and the likelihood of expanding requirements are two possible deciding factors.
When we think of hosted applications, we think of browser or terminal services–based applications. However, hosted applications can be designed using any of these architectures. Everquest is essentially a hosted application for subscribers, but it uses an entirely different model based on an Internet-enabled standalone application.
The bottom line is that there are lot of good choices for developing Internet-enabled applications. The range of options keeps growing rather than becoming more monolithic. Just when it looks like one approach will dominate, different technologies emerge to compete. The trick is picking the right technology for the right application at any given time.